Skip to content

Deploy Your Agent

Get your trained agent to production.

Deployment Options

Local Deployment

iovalence deploy --model model.pkl --target local

Make predictions:

iovalence predict --model model.pkl --input "text here"

Docker Deployment

iovalence deploy --model model.pkl --target docker
docker run -p 5000:5000 my-agent:latest

Cloud Deployment (AWS, GCP, Azure)

iovalence deploy --model model.pkl --target aws
# or --target gcp, --target azure

REST API Server

iovalence serve --model model.pkl --port 5000

API endpoint:

curl -X POST http://localhost:5000/predict \
  -H "Content-Type: application/json" \
  -d '{"input": "your text here"}'

Pre-Deployment Checklist

  • [ ] Model achieves desired accuracy
  • [ ] All dependencies listed
  • [ ] Configuration files included
  • [ ] Data preprocessing documented
  • [ ] Error handling implemented
  • [ ] Security configured
  • [ ] Monitoring enabled

Model Optimization

Quantization (Smaller model)

iovalence quantize --model model.pkl \
  --output model_optimized.pkl \
  --bits 8

Reduces model size by 4x with minimal accuracy loss.

Compression

iovalence compress --model model.pkl \
  --output model_compressed.pkl

Deployment Monitoring

After deployment, monitor:

iovalence monitor --model deployed-model

Metrics: - Request rate - Latency - Error rate - Model accuracy on live data

Version Control

iovalence tag --model model.pkl --version v1.0
iovalence tag --model model.pkl --version v1.1

List versions:

iovalence list-versions --model model.pkl

A/B Testing

Compare two models:

iovalence ab-test \
  --model-a model_v1.pkl \
  --model-b model_v2.pkl \
  --split 0.5

Update Model

When retraining:

iovalence update --model new-model.pkl

Automatic rollback if accuracy drops.

Common Deployment Issues

Issue Solution
Model not found Check path and permissions
Out of memory Use quantization
Slow inference Increase batch processing
High error rate Validate data preprocessing

Best Practices

  1. Always version your models
  2. Test before production
  3. Monitor in production
  4. Have rollback plan
  5. Document everything

Next Steps

Full Deployment Guide →